A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted...
Transcript of A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted...
![Page 1: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/1.jpg)
A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central Valley AquiferKatherine M. Ransom, Bernard T. Nolan, Jon Traum, Claudia C. Faunt, Andrew M. Bell, Jo Ann M. Gronberg, David C. Wheeler, Celia Rosecrans, Bryant Jurgens, Gregory E. Schwarz, Kenneth Belitz, Sandra Eberts, George Kourakos, and Thomas Harter
Nitrate mapping in the Central Valley aquifer
![Page 2: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/2.jpg)
Study Goals and Overview
Rosecrans et al.
To map groundwater nitrate concentration “wall to wall and top to bottom”
Gain understanding of the system
Groundwater age, field scale nitrogen input, oxidation/reduction potential
Boosted Regression Trees
![Page 3: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/3.jpg)
Groundwater Aquifer
Nitrate in Groundwater - Sources
*Nitrogen Cycle image: Modified from University of Wisconsin Integrated Pest and Crop Management, shown on http://fyi.uwex.edu/discoveryfarms/page/6/.
Domestic wastewater is a potential source in rural and urban areas from septic tanks or leaky sewer lines (Bremer and Harter, 2012, and Viers et al., 2012).
Natural sources (organic matter decay) contributes a minimal amount.
![Page 4: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/4.jpg)
Nolan and Hitt, 2006. Vulnerability of shallow groundwater and drinking-water wells to nitrate in the United States,Environmental Science and Technology, 40, 7834-7840.
Nitrate in Groundwater - US
![Page 5: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/5.jpg)
Nitrate in Groundwater – ModelsAuthors Scale Method(s)Nolan, Hitt, and Ruddy, 2002
National Logistic Regression
Nolan and Hitt, 2006 National Non-linear RegressionNolan et al., 2014 Central Valley Logistic Regression,
Random ForestNolan, Fienen, and Lorenz, 2015
Central Valley Boosted Regression Trees, Bayesian Networks, Artificial Neural Networks
Ransom et al., 2017 Central Valley Boosted Regression Trees
Nolan and Hitt, 2006. Vulnerability of Shallow Groundwater and Drinking-Water Wells toNitrate in the United States, Environmental Science and Technology, 40 (24), 7834-7840.
Nolan et al., 2014. Modeling Nitrate at Domestic and Public-Supply Well Depths in theCentral Valley, California, Environmental Science and Technology, 48 (10), 5643-5651.
Nolan, Hitt, and Ruddy, 2002. Probability of Nitrate Contamination of Recently Recharged Groundwaters in the Conterminous United States, Environmental Science and Technology, 36 (10), 2138-2145.
Nolan et al., 2015. A statistical learning framework for groundwater nitrate modelsof the Central Valley, California, USA, Journal of Hydrology, 531, 902-911.
Ransom et al., 2017. A hybrid machine learning model to predict and visualize nitrateconcentration throughout the Central Valley aquifer, California, USA, Science of the Total Environment, 601-602, 1160-1172.
![Page 6: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/6.jpg)
Building on Previous WorkHybrid Approach Oxidation/reduction potential Groundwater age Nitrogen loading – field scale3D map Predictions mapped at depth Interpolation between predictions
![Page 7: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/7.jpg)
Machine Learning for NitratePros Relations need not be linear or follow a particular data
distribution Screens large numbers of variables Handles missing data Results not affected by collinearity Automatically incorporates interactions and thresholds Useful for inferenceCons Overfitting the data Model is harder to interpret Perceived as “black box”
Modified from: B.T. Nolan, 2017
![Page 8: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/8.jpg)
Statistical Methods - Workflow• Predictor variables attributed to wells, 145 total• Boosted regression tree modeling• Predictors ranked based on importance (variable reduction routine)• Top 25 variables kept for final• Predictions made at 17 depths, 3D map created
Measured concentrations
15.24 m deep
30.48 m deep
45.72 m deep
60.96 m deep
![Page 9: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/9.jpg)
Well Data and Predictor Variables
![Page 10: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/10.jpg)
A) Shallow B) Deep
Shallow:1400 wellsDomestic wells180 ft/54.9 m27% exceedance
Deep:2108 wellsPublic wells400 ft/121.9 m6% exceedance
1662 “Hold-out” wells (not shown)
3508 Training wells (shown)
!(
!(
!(
!(!(!(!(
!(
!(
!(
!(
!(!(!(!(
!(!(!(!(!(
!(!(
!(
!(
!( !(!(!(
!(
!(!(
!( !(
!( !(
!(
!(
!(
!(
!(!(!(
!(!(
!(!(
!(
!(
!(
!(!(
!(
!( !(!(
!(!(
!(
!(
!(!(
!(!(
!(!(
!( !(
!(!(
!( !( !(!(
!(!(!( !(!(!(!( !( !(
!(
!(!( !(
!(
!(!(
!(!(
!(!(
!( !(!(!(!(
!(!(!(
!(!(
!(
!(!( !(!(!(
!(!(!(!(!(
!(!(!(
!(!(!( !(
!(
!(!(
!(!(!(!(
!(
!(!(
!( !(!(!( !( !(
!(!(
!(
!(!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(!(!(!(!(
!(
!(
!(!(!(!(
!(!(
!( !(
!( !(
!(!(!(
!(!(
!( !( !(
!(!( !( !(!(
!(
!(
!(!(
!(!(
!(!( !( !(!(
!( !(
!(!(
!( !(!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(!(!(!(
!(
!(!(!(
!(
!(
!(
!(!(!(
!(
!(!(!(!(
!(!(!(!(
!(
!(!(!(
!(!(!(!(
!(!(!(!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(!(!(!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(!(!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!( !(!(
!(
!(
!(
!(!(
!( !(!(!(
!( !(
!(
!(!(!(
!(
!( !(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(!(!(!(
!(!(!(
!(
!(!(
!(!(!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(!(
!(!(
!( !(!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(!(
!(!(!(
!(
!(
!(
!(
!(
!(!( !(
!(!(!( !(
!( !(!(!(!(
!(!(!(!(!(!(
!(!(!(!(!(!(!(
!(!(
!(
!(
!(
!( !(
!(!(
!(
!(
!(
!(!(
!(
!(
!(!(!(!(!(!(
!(
!(!(
!(
!(!(!(!(!(!(!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(!(
!(
!(
!(!(
!(
!(!(
!( !(!(
!(!(
!(!(
!(!(!(!(!(!(!(!(!(
!(
!(!(
!(
!(
!(
!(
!(!(!(!(!(!( !(!(
!(!(
!( !(!( !(
!( !(!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(!(
!(!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(!(
!(!(
!(
!(
!(
!(!(!(!(
!(!(!( !(!(!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(!(
!(!(
!(!(
!(!(
!(
!( !(
!(
!(!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(
!(!(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!( !(!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(!(!(!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(!(
!(
!(
!(!(
!(
!(
!( !(!(!(
!( !(!(!(!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!( !(
!(
!(
!(!(!(!(!(
!(!(!(
!(
!(
!(!(
!(!(
!(!(
!(
!(
!(
!(!(!(
!(
!(!(
!( !(
!(
!(
!(!(
!(!(
!(
!(!(
!(
!(
!(!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(!(
!(
!( !(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(!(
!( !(
!(
!(
!(
!(
!(
!(!(
!( !(
!(!(
!(
!(
!(
!(
!(
!(!( !(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!( !(
!(
!( !(
!(!(!( !(
!(
!(
!(!(
!(
!(
!(
!( !(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(
!(!( !(
!(
!(
!(!(
!( !(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(!(!(
!(
!(!(
!(
!(
!(!(!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(!( !(
!(
!(
!(
!(
!(
!(!(
!( !(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!( !(
!(!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(!(!(
!(!(!(
!(!(!(
!(!(!(
!( !(!(
!(!(
!(!(!(!(
!(!(
!( !( !(!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(!(!(
!(
!(
!(!(!(!(
!(!(!( !(
!(
!(!(!(
!(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(!(
!(!(
!(!(
!(!(
!(
!(
!(!(!(
!(!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(!(
!(!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!( !(
!(
!(!(
!(!(
!(
!(!(!(
!(!(!(!(
!(!(
!(!(
!(!(
!(
!(
!(!(!( !(
!(!(
!(
!(
!(!(!(
!(
!(!(!(
!(
!(!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(!(
!(
!(
!( !(
!( !(!(
!(
!(!( !(
!(
!(!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(
!(
!(
!(
!(
!(
!(!( !(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(!( !(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(!( !(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(!(
!( !(
!(!(
!(!(!(
!(!(!(!(!(!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(!(
!(!(
!(!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(
!( !(
!(!(
!(!(!( !(
!(!(
!(
!(!(
!( !(!(!(!(
!(!(
!(!(!(
!(
!(
!( !( !(!(!(
!(!(
!(!(!(
!( !( !(
!(
!(
!(!(
!( !(
!(
!( !(
!(!(
!( !(!(
!(
!(
!( !(!(!(!(!(
!(
!(
!(
!(!(!(!(!(!(!(
!(
!(!(!(!(
!(
!(
!( !(!(!(!( !(!(
!(
!(
!(
!(
!( !( !(
!(
!( !(
!(
!(!(!(
!(!(!(
!(!(
!(
!( !(!(
!(
!(!(!(
!(!(!(!(!(!(!(!(!(
!(!(
!(
!(!(!(
!(
!(!(!( !(
!(!(
!(!(!( !(!(!( !( !(
!(!(
!(
!(
!(!(!(!(
!(
!(
!(
!(!(
!( !(!(!(
!(!(
!(!(!(!(!(!(
!(
!(!( !(
!(!(
!(!(
!(
!( !(!(
!( !( !(
!(!(
!(
!( !(!(!(!(!(
!(!(!( !(!(!(!(!( !(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!( !(
!(!(!(
!(
!(!(
!(!(!(!(!(!(!(
!(!(!(!( !(
!( !(!(
!(
!(!(
!(!(
!(
!(!(!(!(
!(
!(
!(!(
!(!(!(!(
!(!(
!(!(
!(!( !(
!(!( !(!(!(!(!( !(
!(!(!(
!(!(
!(!(!(!(!(!(
!(!(!( !(!(
!(!(!(!(!( !( !(
!(!(!(
!(
!(!(!(
!(
!(!(!(
!(
!(!(!(!(
!(
!(!(
!(!(
!(!(
!( !(
!(
!( !(
!(
!(
!(!(!(
!( !(!(
!(
!(!(!(
!(!(
!(
!(!(!(!(
!( !(
!(!(!( !(
!( !(!(
!(!(!( !(
!(
!(!(!(
!(
!(
!(
!( !(!(
!(!(!(
!( !(!( !(
!(!(!(!(
!(!(
!(
!(!(
!( !( !(
!(!(
!(
!(
!(!(!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(!(!(
!(
!(
!(
!(
!(!( !(
!(
!(!(!(!(!(!(
!(!(
!(!(!(
!(
!(!(
!(
!( !(
!(!(!(!( !(!(!(
!(
!(
!( !(
!(!(
!(
!(!(
!( !(
!(
!(
!(
!(!(
!(!(
!(
!( !(
!(!( !(!(!(!(!(
!(!(
!(!(!(!(
!(!( !(
!(!(!(!(!(!( !(!(!( !( !(
!(
!(!(!(!(
!(!(!(!( !(
!(!(
!(!( !(!(!(!(
!(
!(
!(
!(
!( !(!(!(
!(!(
!(
!(!(
!( !(
!(
!(
!(!(
!(!(
!(!(!(
!(!(
!(
!(
!(!(
!(!(
!(!(
!(
!(!(
!(!(
!(!(!(
!(
!(!(!(
!(!(!( !(
!(!( !( !(!(!(!( !(
!(!(
!(!( !(
!(!(
!(!(
!(
!(
!(!(
!(!(!(!( !(
!(
!(
!(!(
!(
!(
!(!(!(!(!(
!(
!(
!(
!(
!(!(!(!(!(!(
!(!(
!(
!(!(!(
!(
!(!(!(!(
!(!(
!( !(
!(
!(!(!(
!(!(!(!(!(!(
!(
!(
!(!(
!(
!(!(!( !(!(!(!(!(!(!(!(
!(!(!(!(
!(
!(!(
!(
!(!(!(!(
!(!(
!(
!(
!(!(!(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(!(
!(!(
!(
!(
!( !( !(
!(
!(!(!(!(!(
!(!(!( !( !(
!( !(
!(
!(!(
!(!(!(
!(!(
!(
!(!(
!(!(
!(!(!(!(!(
!(!(
!(!(!(!(
!(
!(!(!(
!(!(
!(!(
!(!( !(!(
!(
!(
!(
!( !(!(!(!(
!(
!(!(!(!(
!(
!(
!(
!(!(!(!(!(!(!(!(!(!(
!(!(!(
!(!(
!(!(!(
!( !(!(
!(!(!(
!(!(!(!(!(!(
!(
!(
!(!(
!(!(!( !(
!(!(!(!(!(
!(
!(!(!(
!(!(
!(!( !(!( !(!(!( !(!(!(
!(!(
!(!(!(!(
!(
!(!(
!(!(
!(!(
!(
!(
!( !(
!(!(
!(
!(!(!(!(!( !(!(!(
!( !(!(
!(
!(
!(
!(!(
!(
!( !(
!(!(!(!(
!(!(!( !(
!(
!( !(
!(
!(!(!(!(!(
!( !(
!(
!(
!(!(!( !(
!(!(!(!(!(!(!(!(!(!(!(
!(!(!(!(!(!(!(
!(!(
!(!(
!(!(
!(!(
!(
!(!(!(!(!(
!(!(!(!( !(!( !(!(
!(
!(
!(!(!(!(!(!(!(
!( !(
!(!(
!( !(!( !(!(!(!(
!(!( !(!(!(!(
!( !(!(
!(!(!(
!(!(!(!(!(!(!(!(
!(!(
!(!(!(!(
!(!(!(!(!(!( !( !(!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!( !(
!(
!(!(
!(
!(!(
!(
!(
!(!(!(
!(
!(!(
!( !(
!(!(
!( !(
!(
!(
!(
!(
!(!(!(
!( !(!(
!(!(
!(!( !(
!(!(
!(
!(
!(!(
!(
!(!(
!(!(!(
!(!( !(
!(
!(!(
!(
!( !(
!(
!(!(
!( !(
!(
!(!(!(
!(!( !(!(
!( !(!(!(
!(
!(!(
!(!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(!(
!(!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(!( !(!(
!( !(!(!(
!(!(
!(
!(!( !(
!(!(
!(!(!(
!( !(!(
!(!(!(
!(!(!(
!(
!(
!(!(
!(!( !( !(
!(!( !(!(
!(
!(!(!(!(!( !(!(!( !(
!(
!( !(!( !(!(!(
!(!(
!( !( !(
!(!(!( !(!(!(!(
!(!(
!(!(!(!(!(!(!(
!(!(!(!(!(!(
!(!(
!(!(!(!(
!(!(!(
!(!(!(!(!(
!(!(!(!(!( !(!(!(!(
!(
!(!(
!(
!(!(!(
!(
!(!( !(
!(!(
!(!(
!(!(!(!( !(!(
!(!(
!(!(!(!(
!(!(!(
!(
!(
!(!(
!(
!(!(!(!(!(
!(!(!(!(!(
!(
!(!(!(!(
!(
!(!(!(
!(!(
!(!(
!(!(
!(!(!(
!(
!(
!(
!(
!(
!(
!(!(!(
!(!(
!(
!(
!(!(
!(!(!(
!(!(!(!(
!(!(!(!(
!(
!(!(!(
!(!(!(!(
!(
!( !(!(!(!(!(!(
!(!(!( !(
!(
!(!(!(!(!(!( !(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(
!(!(
!(!(!(
!(
!(
!(!( !(!(
!(!(!(
!(
!(
!( !(
!(
!(!(
!(!(
!(
!(
!( !(
!(
!(!(
!(!(!(
!(
!(!(!(
!(!(
!(
!(!(
!(
!(!(
!(!(
!(
!(!(
!(!(
!( !(
!(
!(!(!(
!(!(!(
!(!(
!(!(!(
!(!(!(!(!(!(!(!(!(!(
!(
!( !( !(
!(!(!(!(!(
!(!(
!(!(!(
!(!(
!(!(!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(
!(
!(
!( !(
!(!(!(!(
!( !(
!(!(
!(!(
!(
!(!( !(
!(
!(!(
!(
!(!(
!(!( !(
!(!(
!(
!(
!(!( !(
!(!( !(
!( !( !(!(!(!(!(!(
!(!(!( !( !(
!(
!(!(
!(!( !(
!(!(!(!(
!(!( !(
!(!(
!( !(
!(!(!(
!(!(
!(!(!( !(!(
!(!(!(!(
!(!( !( !(!(!(!( !(
!(!(
!(!(!(!(
!(!(
!(!(
!(
!(!(!(
!(!( !(!(
!(!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(!(
!(!(
!( !(!(!(
!(!( !(
!(!(!(!(!(
!(!(
!(!(
!(!(
!(!(
!(
!(!(
!( !(
!(!(!(
!(
!(
!(!(!(
!( !(!(
!(!(!(!(
!( !(
!( !(
!( !(
!(
!(!( !(
!(
!(!(!(!(!(!(
!(!(!(!( !(
!(!( !(
!(!(
!(!(!( !(!(
!(!( !(!(!(
!(!(!(!(!( !(
!(
!(!(!(
!(!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!(
!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(!(
!(
!(!(!(
!(!(!(
!(
!(!(
!(
!(!(!(
!( !(
!(!(!(
!(
!(!(!(!(!(
!(!(
!(
!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(!(!(
!(!(!(!(!(
!(
!(
!(!(!(
!(!( !( !(!(!(
!(
!(!(
!(
!(!(
!(
!(
!(!( !(
!(
!(
!(
!(
!(
!(!(!(
!( !(!(!(
!(
!(
!(
!(
!( !(
!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(
!(!( !(
!(
!(
!(
!(
!(
!(!(
!( !(
!(!(
!(
!(
!(
!(!(!(
!(
!(
!(
!(
!(!(
!(
!(
!(!(
!(!(!(
!( !(
!(
!(
!(
!(!(
!(!(
!(!(!(!(
!( !(
!(!( !(
!(!(!( !(!( !(
!(
!(!(
!(
!(!(!(
!(
!(
!(!(
!(!(
!(
!(!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(!(
!(
!(
!(
!(!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(
!(!(
!(!( !(
!( !( !(!(
!(!(
!(!(!( !(!(!(
!(
!(
!(!( !( !(
!(!(!(
!(
!(!(
!( !(
!(
!(!(
!(
!(
!(!( !(
!(!(
!(
!(
!(
!(!(
!(!(!(
!(!(!(!( !(!(
!(!(
!(!(!( !(!(
!(
!( !(
!(
!(!(
!(!(
!(
!(
!(
!(!(
!(
!(
!(
!(
!(
!(
!(
!(
!( !(
!(
!(
!(!(
!(
!(
!(
!(!(
!(
!( !(
!(
!(
!(
!(
!(
!(
B) DeepA) ShallowCALIFORNIA
EXPLANATIONNitrate concentrationin groundwater,in milligrams per liter, as N
!( 0 to 2
!( >2 to 4
!( >4 to 6
!( >6 to 8
!( >8 to 10
!( >10
0 50 100 Miles
0 40 80 Kilometers
East Fans
West Fans
Basin
SacramentoValley
San JoaquinValley
![Page 11: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/11.jpg)
Probability of Anoxic ConditionCALIFORNIA
EXPLANATION
Probability of DO < 0.5 ppm< 0.15
0.15 - 0.3
0.3 - 0.45
0.45 - 0.6
0.6 - 0.75
> 0.75
0 50 100 Miles
0 50 100 Kilometers
Domestic well depth Public supply well depth
![Page 12: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/12.jpg)
• Key component not included in previous models.
• “Proxies” such as well depth or depth to water.
MODFLOW/MODPATH Estimates of Groundwater Age with Depth
Estimates from: Central Valley Hydrologic Model, Faunt, C. C. (2009). Groundwater availability of the Central Valley Aquifer, California. Professional Paper 1766, U.S. Geological Survey.
![Page 13: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/13.jpg)
Field-Scale Nitrogen Leaching Flux - 1975
Based on nearly 200 land use types, including 60 crop types.
Available for 1945, 1960, 1975, 1990, and 2005.
CALIFORNIA
EXPLANATION
Unsaturated zone nitrogenleaching flux togroundwater, 1975
< 4
4 - 6
6 - 8
8 - 10
> 10
0 50 100 Miles
0 40 80 Kilometers
![Page 14: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/14.jpg)
County-Scale Nitrogen InputCALIFORNIA
EXPLANATION
Total landscape nitrogen input,1992 (kg)
<=2000
>2000 - 4000
>4000-6000
>6000-8000
>8000-10000
>10000
0 50 100 Miles
0 40 80 Kilometers
![Page 15: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/15.jpg)
Statistical Methods - Software
• caret• gbm• raster• sensitivity• boot
Modeling and Prediction
Packages
Variable Processing 3D Visualization
![Page 16: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/16.jpg)
Statistical Methods - Boosted Regression Trees• aka Gradient Boosting Machine• An ensemble method: collection of many small models (boosting) • Based on classification trees • Each new tree built on the residuals of the previous tree (gradient)• Randomness added by subsampling data • Trees controlled by tuning aka metaparameters
Simple Regression TreeExample Apartments Dataset
![Page 17: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/17.jpg)
Results – Model Performance
Training RMSE: 0.705Training R2: 0.825
Hold-out RMSE: 1.132Hold-out R2: 0.443
Residual Comparison
![Page 18: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/18.jpg)
• To 1600 ft below ground surface• 17 predicted layers• Linear interpolation• 1 m vertical resolution
Results – Oasis Montaj 3D map
![Page 19: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/19.jpg)
Results – Predictions at Specified DepthsPrivate well depth Public supply well depthCALIFORNIA
0 180 360 Miles
0 160 320 Kilometers
EXPLANATION
Nitrate - N (mg/L)
< 2
2 - 4
4 - 6
6 - 8
8 - 10
> 10
West Fans
Basin
East Fans
![Page 20: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/20.jpg)
Secondary Results - Importance Ranking
![Page 21: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/21.jpg)
Secondary Results – Partial Dependency Plots
Probability of Anoxic Conditions - DO Probability of Anoxic Conditions - Mn
LN(N
O3-
N) m
g/L
LN(N
O3-
N) m
g/L
Probability of dissolved oxygen < 0.5 ppm Probability of manganese > 50 ppb
![Page 22: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/22.jpg)
Secondary Results – Partial Dependency Plots
Natural and Water Land Use, 1990sDistance to River
LN(N
O3-
N) m
g/L
LN(N
O3-
N) m
g/L
Distance to river with stream order > 3, m Area surrounding well as natural land use, m2
![Page 23: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/23.jpg)
Secondary Results – Partial Dependency PlotsLN
(NO
3-N
) mg/
L
LN(N
O3-
N) m
g/L
Natural and Water Land UseProb of DO < 0.5 ppm
LN(N
O3-
N) m
g/L
Natural and Water Land UseProb of Mn > 50 ppb
![Page 24: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/24.jpg)
Summary and Conclusions• Mapped nitrate tended to decrease with depth• Alluvial fans region had higher nitrate concentrations than basin subregion• Anoxic conditions highly related to nitrate concentration• Patterns on partial plots make intuitive sense• Coming soon: updated national nitrate and arsenic maps
Locating High Risk Domestic Wells• Cookie cutter national models (updated or current) for full coverage• Use estimates from current national arsenic model (Ayotte et al., 2017)• Develop new California specific model• Consider multiple constituents together (multinominal BRT)?• Nitrate, arsenic, uranium, others?• Overlay with well locations
Arsenic Nitrate Clean
Soil Type A Soil Type B
Prob Anoxic < 0.6 Prob Anoxic > 0.6
Reference: Estimating the High-Arsenic Domestic-Well Population in the Conterminous United States, Ayotte et al., Environmental Science and Technology , 2017, 51 (21) pg. 12442 – 12454. https://pubs.acs.org/doi/10.1021/acs.est.7b02881
![Page 25: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/25.jpg)
Questions?
Article available at:https://www.sciencedirect.com/science/article/pii/S0048969717313013?via%3Dihub
Data raster grids available at:https://www.sciencebase.gov/catalog/item/58c1d920e4b014cc3a3d3b63
![Page 26: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/26.jpg)
Appendix
![Page 27: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/27.jpg)
Statistical Methods – Cross ValidationMetaparameters:interaction depth, shrinkage, number of trees, size of terminal nodes
CV tuning addresses over fit by limiting model complexity
Credit: Hastie et al., 2009. The Elements of Statistical Learning.
![Page 28: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/28.jpg)
Statistical Methods - Variable Reduction
1 % difference
Increase in Prediction Errors to Hold-out Data
![Page 29: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/29.jpg)
Results – Prediction Intervals
199 models made with bootstrappedsets of the training data
199 predictions made to hold-outdata
![Page 30: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/30.jpg)
Results – Prediction Interval WidthPrivate well depth Public well depthCALIFORNIA
0 180 360 Miles
0 160 320 Kilometers
EXPLANATIONRelative prediction interval width
< 4
4 - 8
8 -12
> 12
West Fans
Basin
East Fans
![Page 31: A Hybrid Boosted Regression Tree Model to Predict and ......Jan 18, 2019 · A Hybrid Boosted Regression Tree Model to Predict and Visualize Nitrate Concentration Throughout the Central](https://reader030.fdocuments.in/reader030/viewer/2022040907/5e7d241d76e33356b56c28fd/html5/thumbnails/31.jpg)
Results – Sobol Sensitivity Indices